Analysis of the co-evolutions of correlations as a tool for QSAR-modeling of carcinogenicity: An unexpected good prediction based on a model that seems untrustworthy

Alla P. Toropova, Andrey A. Toropov, Rodolfo Gonella Diaza, Emilio Benfenati, Guesippina Gini

Research output: Contribution to journalArticle

Abstract

To validate QSAR models an external test set is increasingly used. However the definition of the compounds for the test set is still debated. We studied, co-evolutions of correlations between optimal descriptors and carcinogenicity (pTD50) for the subtraining, calibration, and test set. Weak correlations for the sub-training set are sometimes accompanied by quite good correlations for the external test set. This can be explained in terms of the probability theory and can help define a suitable test set. The simplified molecular input line entry system (SMILES) was used to represent the molecular structure. Correlation weights for calculating the optimal descriptors are related to fragments of the SMILES. The statistical quality of the model is: n=170, r2=0.6638, q2=0.6554, s=0.828, F=331 (sub-training set); n=170, r 2=0.6609, r2 pred=0.6520, s=0.825, F=331 (calibration set); and n=61, r2=0.7796, r2 pred=0.7658, Rm 2=0.7448, s=0.563, F=221 (test set). The calculations were done with CORAL software (http://www.insilico.eu/ coral/).

Original languageEnglish
Pages (from-to)165-174
Number of pages10
JournalCentral European Journal of Chemistry
Volume9
Issue number1
DOIs
Publication statusPublished - 2011

Keywords

  • Balance of correlations
  • Carcinogenicity
  • Co-evolutions of correlations
  • Optimal descriptor
  • QSAR

ASJC Scopus subject areas

  • Chemistry(all)
  • Materials Chemistry

Fingerprint Dive into the research topics of 'Analysis of the co-evolutions of correlations as a tool for QSAR-modeling of carcinogenicity: An unexpected good prediction based on a model that seems untrustworthy'. Together they form a unique fingerprint.

  • Cite this